Thierry Grenot January 25, 2024

# Conversational access to business software: 10 criteria for the product director

## Context and evolution

ChatGPT’s public release in December 2022 highlighted the use of natural language for complex searches. Business software user interfaces are reaching a “tipping point” with three generations of interaction:

1. **The Web** — technological development of the 1990s (Internet, JavaScript, XML)
2. **Mobile applications** — arrival of iPhones with app stores (2007)
3. **Conversational access** — direct expression of requests through simple conversation

For occasional users or frontline workers, conversation is the most obvious access method. The benefits are multiple: improved user experience quality, increased adoption rate, reduced customer service burden, and innovative market positioning.

## The 10 essential criteria

### 1. Use cases

Identify functions without complex graphical interfaces. Complete products (HRIS, ERP, Finance) typically offer “a few dozen relevant use cases” for a conversational interface.

### 2. Flexibility

Adapt the implementation to future evolutions without excessive effort. Handle business jargon, proper nouns and authentication.

### 3. Access channels

Diversify presence across MS Teams, Slack, WhatsApp, mobile apps and even voice, to maximise access while offering “a unified experience”.

### 4. Multilingual

Ability to offer conversational access simultaneously to the entire customer base and to add new languages easily.

### 5. Implementation timeline

“Unless your team already has sharp NLP (Natural Language Processing) skills, don’t expect quick results.” Estimated timeline: 6 to 24 months for full rollout.

### 6. Cost

- **Design and development:** €60,000 to €600,000  
- **Operations:** NLP and generative AI tool costs  
- **Maintenance:** necessary and non-negligible

### 7. Available skills

Balance corrective maintenance against innovation. Focus efforts on the core business and user quality.

### 8. User experience

Essential criteria:

- Comprehension quality  
- Tolerance for input errors  
- Response quality  
- Handling of incorrect requests  
- User assistance  
- Absence of hallucinations  
- Exchange speed

### 9. Explainability

NLP is not magic and can be opaque. Risks: bias, hallucinations, fake news. Mastering the technologies and data enables the required transparency.

### 10. Security, privacy, sovereignty

“Business applications often handle confidential, even personal data.” A “security & privacy by design” architecture is required.

## Conclusion

Technologies now make conversational interfaces accessible to all business software publishers, but they require “finesse, expertise and attention”, as they become a major point of contact with customers.